Originally posted by: DSL

Hi everyone,

I'm using CPLEX 12.5 Concert C++. In our application, we have to solve a sequence of similar LP models. That is, we construct an initial LP_1, extract and solve it. Afterwards, some RHS coefficients of LP_1 need to be modified resulting in LP_2. Then, LP_2 is again extracted and solved. Modifying LP_2 yields LP_3 and so on.

The question arising is whether CPLEX automatically uses any information that is available after LP_i is solved to warm start the solution process of LP_i+1.

Kind regards,

David

Originally posted by: RWunderling

David,

you should not re-extract the model after you have manipulated LP_1 to get LP_2.  In this case CPLEX will maintain the optimal basis from LP_1 and use it as starting basis when solving LP_2. By reextracting the model, CPLEX will assume you start a new problem and discard all information.

Roland

Originally posted by: DSL

Perfect, thanks!

I assume that cplex shows the same behaviour if MIPs are being solved instead of LPs.

Originally posted by: RWunderling

Yes, but you have to keep in mind that restarts for MIP are much less powerfull.  Essentially, all that can be saved are previous solutions.  Those are checked if they give rise to new solutions to the modified problem.  But almost everything else needs to be computed from scratch as it is generally incompatible with problem manipulations.

Roland